zeroGPU v1

modules/pmbl.py

@@ -1,6 +1,8 @@
 import sqlite3
 from datetime import datetime
 from concurrent.futures import ThreadPoolExecutor
+import torch
+from huggingface_hub import spaces
 
 class PMBL:
     def __init__(self, model_path):
@@ -100,9 +102,11 @@ class PMBL:
 
         for chunk in response.result():
             yield chunk
-
+            
+    @spaces.gpu
     def generate_response_task(self, system_prompt, prompt, n_ctx):
         llm = Llama(model_path=self.model_path, n_ctx=n_ctx, n_threads=8, n_gpu_layers=-1, mlock=True)
+        llm = llm.to("cuda")  # Move the model to the GPU
 
         response = llm(
             system_prompt,
@@ -146,9 +150,11 @@ class PMBL:
             conn.commit()
 
         conn.close()
-
+        
+    @spaces.gpu
     def generate_topic(self, prompt, response):
         llm = Llama(model_path=self.model_path, n_ctx=1690, n_threads=8, n_gpu_layers=-1, mlock=True)
+        llm = llm.to("cuda")  # Move the model to the GPU
 
         system_prompt = f"Based on the following interaction between a user and an AI assistant, generate a concise topic for the conversation in 2-4 words:\n\nUser: {prompt}\nAssistant: {response}\n\nTopic:"